Protectionism and the Education–Fertility Tradeoff in Late 19th
century France∗
Cecilia Garcı́a-Peñalosa
Aix-Marseille University and CESifo‡
Vincent Bignon
Banque de France†
February 13, 2015
Abstract
The assumption that education and fertility are endogenous decisions that react to economic circumstances is a cornerstone of the unified growth theory that explains the transition
to modern economic growth, yet evidence that such a mechanism was in operation before
the 20th century is limited. This paper provides evidence of how protectionism reversed the
education and fertility trends that were well under way in late 19th-century France. The
Méline tariff, a tariff on cereals introduced in 1892, led to a substantial increase in agricultural wages, thus reducing the relative return to education. Since the importance of cereal
production varied across regions, we use these differences to estimate the impact of the tariff.
Our findings indicate that the tariff reduced education and increased fertility. The magnitude of these effects was substantial, and in regions with large shares of employment in cereal
production the tariff offset the time trend in education for up to 15 years. Our results thus
indicate that even in the 19th century, policies that changed the economic prospects of their
offspring affected parents’ decisions about the quantity and quality of children.
JEL Classification: J13, N33, O15
Key words: Education, fertility, unified growth theory, France.
∗
For discussions and suggestions, we thank Eve Caroli. The opinions expressed here are not necessarily those
of the Banque de France or the Eurosystem.
†
DGEI-DEMFI-Pomone 41-1422, 31 rue Croix des Petits Champs, 75001 Paris cedex, France - Email :
[email protected]
‡
GREQAM, Centre de la Vieille Charité, 2 rue de la Charité, 13002 Marseille - Email : [email protected]
1
1
Introduction
The causes of the emergence of modern growth remain hotly debated amongst economists. One
of the most influential theories is unified growth theory (from now onwards, UGT), developed
by Galor and Weil (1999), Galor and Weil (2000) and Galor and Moav (2002) which proposes
a mechanism through which economies move endogenously from a s subsistence Malthusian
economy into a regime with growing per capita incomes. UGT builds on two key elements. On
the one hand, technological change depends on population size and the level of education of
the labour force. On the other, population growth and education are determined by household
choices which respond to economic incentives. This second element implies a trade-off between
the quantity and the quality of children that an individual has, with parents choosing between
numerous but little educated children or a few well-schooled offspring. Critics of UGT argue that
it is unlikely that in the 19th century fertility and education were the outcome of rational choices,
and that they were more likely to be shaped by social norms than by economic constraints.1
The aim of this paper is to provide evidence for the fact that such a trade-off was affected by
economic shocks well before the postwar period on which most of the literature has focussed.
Our identification strategy relies on a major policy shock that occurred in France at the
end of the 19th century. Following a massive increase in cereal exports that were arriving to
Western Europe from the America and Russia, cereal prices in France plunged, resulting in
a major income loss for cereal producers (Golob 1944). As was the case in other European
countries, political pressure to impose tariffs on cereal imports grew in the 1880s and led to
the adoption in 1892 of the so called Méline tariff, a tariff that halted the fall in cereal prices
and led to substantial wage increases (O’Rourke 1997). We argue that, under the assumption
that human capital is less productive in agriculture than in manufacturing, the tariff reduced
the relative return to education and, as predicted by UGT, led to a reduction in human capital
investments and an increase in fertility.
We construct a simple model that captures the quantity-quality trade-off. Our economy has
two sectors, agriculture and manufacturing, and we suppose that human capital is productive
only in the latter. Parents derive utility from both the number of children that they have
and from the expected income of their offspring, which generates the usual trade-off between
fertility and investment in children’s education. The latter is in turn determined by the relative
return to education, that is, by the wage in manufacturing relative to that in agriculture and
by the probabilities of being employed in one or the other sector. A tariff on agricultural goods
1
See Guinnane (2011) for a discussion.
2
increases wages in farming and the employment share of the sector, thus reducing the return
to education and leading to lower investments in human capital. Because parents spend fewer
resources in children’s quality, they respond by increasing their quantity, and the tariff results
in higher fertility rates. The larger the initial share of employment in agriculture, the stronger
these effects are since the price increase implied by the tariff represents a large shock to the local
economy.
To take the model to the data we use France’s division into administrative districts. In
the late 19th century, these districts differed greatly in the importance that agriculture, and
in particular cereal production, had in the local economy. We hence construct a measure of
employment in cereal production as a share of total employment for 1892 and interact it with a
dummy taking the value one whenever the Méline tariff was in operation. We then examine the
effect of the tariff on birth rates and fertility rates, and find a negative impact of the dummy
interacted with cereal employment shares which is consistent with the theory. Education is
measured in a variety of ways. We consider enrolment rates in primary education, which at
the time was supposed to cater for children aged between 6 and 13. An important feature
of the period is that the fraction of students older than 13 enrolled in primary education was
substantial. We hence compute enrolment rates for all students as well as for two groups, those
aged 6-13 and those aged 14 and over. For the former group enrolment rates were negatively
affected by the tariff, while for those who are 14 and over we find a positive effect on enrolment
during the decade following the introduction of the tariff. Our model shows that for individuals
whose parents had already completed their fertility and education decisions, the positive income
effect generated by the tariff resulted in an increase in their education investment as young
adults. In our data this effect is, however, moderate and overall enrolment rates fell in response
to the Méline tariff.
We also consider completed education by using data on army conscripts. The period for
which we have data is not long, and spans individuals that were aged between 10 and 20 years
of age when the tariff was introduced. Our results indicate that for older cohorts the tariff had a
positive impact on education, becoming insignificant and then negative as we consider younger
and younger generations. These results provide evidence of a trade-off between education and
fertility in the 19th century as well as supporting theories that maintain that these two variables
react rapidly to economic incentives.
The paper contributes to the literature concerned with identifying the determinants of
parental choices between fertility and education. The model introduced by Becker (1960) and
3
enriched by Becker and Tomes (1976) has been the subject of numerous empirical tests. Most
of this literature has used contemporary data and a variety of identification strategies, such
as considering the impact of the arrival of twins in a household on subsequent education investments; see Rosenzweig and Wolpin (1980) or Rosenzweig and Zhang (2009) for more recent
data. Broadly speaking, the evidence supports the existence of such a trade-off in the second
half of the 20th century, although some results are less supportive (notably Black, Devereux, and
Salvanes (2005) who argue that the impact of family size on education is in fact a relationship
between birth order and education).
In contrast to the numerous studies on a recent data, historical evidence on this trade-off
is scarce, the exceptions being Becker, Cinnirella, and Woessmann (2010) and Bleakley and
Lange (2009). Our analysis shares much with these two papers. The former identify the qualityquantity trade-off using data for 19th century Prussia. They find suitable instruments for
regional differences in education and fertility (sex ratios and distance to Wittenberg) and can
hence identify the impact of one variable on the other. However, in contrast to this paper, their
analysis does not allow them to examine how economic variables affect this trade-off. Our work
is particularly close to Bleakley and Lange (2009) who use disease eradication in the south of
the US around 1910 to analyse fertility and education responses. The exogenous campaigns to
eradicate hookworm, a parasite that particularly affects children’s health, reduced the “price
of child quality” and thus increased the return to human capital. As a result, educational
investments rose and fertility rates fell. We follow a similar empirical strategy by focusing on
the relative return to education. In contrast to Bleakley and Lange (2009) the external shock we
consider has a less direct impact on children’s welfare and rather acts by changing equilibrium
prices and quantities in the economy. What makes the strength Bleakley and Lange’s paper is
also its drawback. Because it relies on a shock that has a direct impact on children’s quality,
the mechanism in operation is well identified, yet it does not provide evidence that aggregate
macroeconomic features impact fertility and education as advocated by UGT. Our analysis
focuses precisely on a major aggregate shock and identifies its consequences for fertility and
education.
The paper is also related to a vast body of evidence trying to identify the determinants of the
demographic transition; see Easterlin (1976) for a discussion. Although we are not particularly
concerned with this episode, since France had the world’s earliest demographic transition which
took place almost a century before the Méline tariff was introduced,2 some of this literature
2
See Chaunu (1972), Van de Walle (1980), Weir (1984) and Bardet and Le Bras (1988) for evidence.
4
proposes an approach closely related to ours by trying to identify variables that affect the cost
of having children. Notably, Schultz (1985) argues that the fertility transition in Sweden, which
took place in the 1880s, was largely the result of changes in international agricultural prices that
raised the relative wage in female-intensive occupations. Exploiting differences across Swedish
counties in the intensity of these activities, he finds that the increase in relative female wages
explains a substantial fraction of fertility changes. Our paper shares with this work its emphasis
on how terms of trade shocks that affect relative wages in a country can lead to rapid fertility
responses.
Lastly, the paper is related to the economic history literature documenting the impact of
late 19th century protectionist policy on economic outcomes. Following Bairoch (1972), numerous studies have found that protectionism was associated with higher growth rates and,
when systematized to a panel of countries, this positive association between growth and tariffs
has generated the so-called tariff-growth paradox; see O’Rourke (1997), O’Rourke (2000), Jacks
(2006). Here we take a different approach; rather than exploiting cross-country differences, we
document that within France the districts that benefited the most from the tariff were also those
where it had the strongest negative effect on children’s education.3
The paper is organised as follows. Section 2 gives the historical background of our study
in terms of agricultural protectionism, education decisions and fertility. Section 3 solves a twosector model of the joint family decision between the number of children and education. Section
4 presents the econometric specification we use to bring the model to the data. The next two
sections present the data and the empirical results. Section 7 concludes.
2
2.1
Historical background
The Méline tariff and its consequence on the economy
The signing of the 1860 free-trade treaty with England has been viewed as a milestone in the
historiography of French attitudes towards international trade (Bairoch 1972). Recent research
argues that economic forces largely anticipated trade politics, see Accominotti and Flandreau
(2008), and Tena-Junguito, Lampe, and Tâmega Fernandes (2012). Nye (2007) shows that effective tariff duties on imports were low in France throughout the century, especially on agricultural
products. The invention of the steamship and the development of the domestic railway network
triggered a decrease of freight rates, especially across the Atlantic (North 1958 and Harley 1988)
that increased grain market integration, see Federico and Persson (2007) and Uebele (2011). The
3
Dormois (2009) uses industry-level data to document the negative impact of industrial tariffs on European
industry.
5
resulting boom in trade was mainly driven by large exports of grains and other primary products from Latin America to Europe which resulted in deflationary pressure on prices in France;
see Kindleberger (1950). Agricultural prices declined more than other prices, thus reducing
farmers’ revenues. Generalised discontent led farmers to lobby for protection but, because of
the alliance between free-traders and industrialists, no majority was obtained in Parliament to
impose protective tariffs .4
The 1889 parliamentary elections changed the population of lawmakers towards a majority in
favor of more protection. Negotiations with the governments and discussions in Parliament lead
to the proposal of an increase in the tariffs on cereals to fight the competition coming from the
Americas.5 Tariffs were introduced ad valorem: for each 100 kilos of cereals, the tariff increased
the import price by 5 francs in 1892, or by about 25% (Golob, 1944, p. 204). The economic
magnitude of the tariff was substantial. Levasseur (1911, vol. II, p. 585) estimates that the
Méline tariff, if applied earlier, would have increased the cereal prices in 1889 by 80%. Moreover,
the law allowed for the tariff to be adjusted every year to take into account variations in the
world price of cereals. According to Augé-Laribé (1950, p. 246-7) (1944, p. 234) there were
thirty major legislative modifications of the tariff structure of 20 years. For example, in 1894
the wheat duty was increased from 5 to 7 francs per hundred kilograms. These changes induced
an upward trend in agricultural duties during the twenty years that followed the adoption of the
Méline tariff.
The magnitude of the effects of the tariff was enormous. In a context in which world price of
grains decreased by a third, O’Rourke (1997) documents that the Méline tariff protected farmers
revenue from most of this decline by increasing domestic prices by 26.5%. In a country in which
the agricultural population represented 50% of the working population (Golob, 1944, p. 18),
the tariff implied that actual French grain output was twice as large as it would have been in
the absence of protection (O’Rourke, 1997, p. 793). The overall effect of the reduction in world
prices and tariffs resulted in a increase of real wages for the entire population but particularly
for farmers, who were made better compared to the rest of the population.
4
In the 1880s the only success of the agricultural lobby in the parliament was the exclusion of agricultural
products from the most-favored nation clause
5
The promoter of the agricultural tariff Méline justified the increase of the tariff by saying to lawmakers that
”suddenly came the development of the means of transportation and communication, the rapid decrease in freight
costs, in a few years placing these great markets [i.e. America, India and Australia] at our door” quoted in Golob
(1944, p.182)
6
50
per 1,000
40
30
20
10
1750
1800
1850
1900
1950
2000
year
The grey area is the treatment period (1892-1913)
Figure 1: The birth rate in France 1740-2013
2.2
Education, fertility and the demographic transition in France
As it is widely acknowledged, France was the first country to experience a fertility transition;
see Guinnane (2011) for a discussion an international context. Figure 1 depicts the crude birth
rate in France over the period 1740 to 2012, with our period of interest (1872-1913) shaded.6
The first few years in the sample exhibit the usual pre-transition birth rate of around 40 children
per thousand individuals. Birth rates started to decline around 1790, almost one century before
the fertility transition took place in England and Germany. The reasons for this early transition
are still poorly understood. It has been argued that the unique and spectacular reduction in
mortality that took place in France in the second half of the 18th century could be a trigger, while
other authors have emphasised the role of wealth and the changes in inequality that followed
the French Revolution; see Guinnane (2011), Cummins (2013), Wrigley (1985a, 1985b) amongst
others. In contrast to other countries, where the late 19th century witnessed major changes in
fertility behaviour, the period just before the introduction of the Méline tariff consists of two
decades of substantial stability, as birth rates in France continued the long- run trend, as can be
6
Blayo and Henry (1975) is the source of the series before 1800. The 1946 INSEE statistical yearbook gives
19th century numbers, with the corrections proposed in Dupaquier (1988). The digitized series on the INSEE
website are the source of 20th century numbers.
7
110
14.5
14
13
12.5
90
Per 1,000 females
as % of population
100
13.5
80
70
1875
1885
1895
year
Enrolment rate
1905
1915
Fertility rate
The vertical line is the year of passing of Meline tariffs
Figure 2: The literacy and the fertility rate, France 1872-1913
seen from figure 1. There is nevertheless a slowdown of the trend after 18 and 92. The birth rate
fell by 2.5 children between 1872 and 1882 and by 1.9 children in the next decade (reductions
of 1 and 0.75%, respectively), yet in the decade following the introduction of the tariff the birth
rate declined by on 0.7 children (i.e. by 0.3%). The fertility changed momentum after the War,
falling by 2.5 children between 1924 (the year in which fertility returned to its pre-war level)
and 1934.
Figure 2 uses our district-level data to compute national aggregates for fertility rates and
literacy of males at age 20. The change in the fertility trend is apparent here. Fertility is
stagnant in the early 1890s and actually increases between1895 and 1901 before declining again.
These increase implied a “delay” in the reduction of fertility of 11 years (the historical minimum
of 1895 being attained again in 1906).
The expansion of education in France took place in the middle of the 19th century, the
result of major legal changes and a substantial investment in education infrastructure; see Prost
1994. Historians of education described the period 1837-1867 as a period of “universalization” of
primary education on by 1881 the majority of districts had attained enrolment rates on hundred
percent. A puzzle in the literature is that of the “lost decade”. Between 1886 and 1896 not
8
only there is no progress in primary schooling, but mainly districts experienced a decline in
enrolment rates, with the average falling by 3.9% and 4.4% for boys and girls respectively; see
Prost 1994, p 71. The timing of these changes raises the question of whether the Méline tariff
was while of the factors behind them.
Figure 2 presents one of our measures of education, which uses army data to compute literacy
rates amongst 20-year old males (see below for the details). Literacy increases over the period,
although there seems to be a substantial slowdown of the trend in the early 1900s. Because
the data considers completed education, the effect of the tariff will not be immediate and it
is not obvious which generation of conscripts will experience a reduction in their educational
attainment. Our econometric analysis below will address this issue.
3
Modelling the quantity-quality trade-off
We consider a two-sector version of the quantity-quality trade-off model developed by Galor and
Weil (1999) and Galor and Weil (2000), although we abstract from technological change. The
production side of the economy features two goods, an agricultural good and a manufacturing
good. The later is the numeraire, while the agricultural good is traded and has an exogenously
given price pt that will be the source of the shock we consider. As in the original model, the
key decision is the choice by households of the number of children and their education, i.e their
quantity and quality, in response to economic incentives.
3.1
Technologies and preferences
The economy produces two goods, an agricultural good and a manufacturing good. The former
is produced using land T and labour Lat according to the following technology
Yat = (AT )1−α Lαat ,
(1)
where Yat is agricultural output, A is agricultural productivity, and 0 < α < 1. The manufacturing good is also produced through a Cobb-Douglas technology of the form
Ymt = B 1−α (ht Lmt )α ,
(2)
where Ymt is manufacturing output, B is a fixed factor in the sector (potentially capital, but we
abstract from its accumulation), ht the average human capital of workers and Lmt employment
in the sector. The price of the manufacturing good is 1, while that of the agricultural good is
9
pt and will be the source of the shock we consider. A crucial assumption in the model is that
human capital increases productivity in the manufacturing sector but not in agriculture.7
The two sectors pay workers their marginal product, and in the appendix we derive the
agricultural wage, wat , and the wage per efficiency unit of labour in manufacturing, wmt . Under
our assumption that education has no impact on agricultural productivity the income of a farmer
is simply wat . In contrast, human capital increases manufacturing productivity, implying that
an agent with ht efficiency units of labour receives a potential income of ht wmt . Labour market
equilibrium requires the equalization of incomes across sectors, i.e. wat = ht wmt , and yields the
fraction of the population employed in agriculture qt and that employed in manufacturing 1 − qt .
We turn next to households’ preferences and constraints. An important aspect in the data
is that the education decisions concerning an individual occur through several years. We hence
assume that agents live for 3 periods. In the first they are born and receive education from their
parents, in the second they are young adults and are endowed with 0.5 units of time. They may
work for the entire period or acquire an amount of education xt at a cost τ x . In the second
period, as mature adults, they are also endowed with 0.5 units of time which they may spend
working or raising children. Borrowing across period is assumed not to be possible.
We suppose that the utility of an agent born at time t is given by
Ut = ct1−γ (nt+2 Eyy,t+2 )γ ,
(3)
where ct is the lifetime consumption of the individual, nt+2 the number of children she has
(which are born at t + 2) and Eyt+2 the expected (potential) income that her offspring (born
at t + 2) will get when she is a young adult. The time cost of bearing nt children is given by
τ q nt , while τ e et nt is the time cost of giving them a level of education et . The budget constraint
is then given by
ct = yy,t (0.5 − τ x xt ) + ym,t (0.5 − (τ q + τ e et+2 )nt+2 ),
where yy,t is the potential income when the individual is a young adult and ym,t that when she
is mature. We suppose that a constant fraction of consumption is allocated to the agricultural
good and the rest to the manufacturing good.8
Young adults whose parents invested et in their education have a level of human capital h(et )
with
h(et ) = βeθt ,
7
(4)
This assumption has support in the French case, see Golob, 1944
It would be straight forward to derive such a result from a Cobb-Douglas utility function. We abstract from
such decision in order to concentrate on the key aspects of the model.
8
10
where β > 0 and θ ∈ (0, 1), implying that h(et ) is increasing in et and exhibits diminishing
returns to the education investment. If the individual then invests in education when she is a
young adult, her human capital becomes h(et , xt ) = βeθt (1 + bxt ), where b > 0. When taking
the education decision of their children, parents suppose that with probability q they will work
in agriculture and with probability (1 − q) in manufacturing. The resulting expected income of
a young adult born at t is
Eyy,t = qt+1 wat+1 + (1 − qt+1 )h(et )wmt+1 .
Clearly, the higher the agricultural wage and agricultural employment are, the lower the relative
return to education will be, thus reducing the incentive of parents to forgo consumption in order
to increase the education of their children. This mechanism will drive our results.
3.2
Solving the model
Education and fertility
The individual’s problem is given by
maxUt = c1−γ
(nt+2 Eyt+2 )γ
t
c,n,e
s.t.
(5)
Eyy,t+2 = qt+3 wat+3 + (1 − qt+3 )h(et+2 )wmt+3
ct = yy,t (0.5 − τ x xt ) + ym,t (0.5 − (τ q + τ e et+2 )nt+2 )
h(et+2 ) = βeθt+2
yy,t = φwat+1 + (1 − φ)wmt+1 βeθt
ym,t = φwat+2 + (1 − φ)wmt+2 βeθt (1 + bxt )
xt ≥ 0, and yy,t (0.5 − τ x xt ) ≥ 0,
et ≥ 0, nt ≥ 0, and 0.5 − (τ q + τ e et+2 )nt+2 ≥ 0.
The first three constraints give the expected income of offspring when young, the consumption
of the individual, and the human capital accumulation function for the offspring. The next two
constraints give the income of the individual in the two periods, where φ is an indicator variable
taking the value 1 if the individual works in agriculture and 0 if he works in manufacturing.
The last two lines give the constraints that fertility, education investments, and consumption in
each period be non-negative.
The consumer’s problem is solved in the appendix, where we suppose that α = 0.5 in order
to get explicit analytical solutions. There we show that, under the assumption that the time
11
cost of young adults’ education is sufficiently high, i.e.
τ x > 0.5b,
then, whenever wages are constant over time, there is no adult education and xt = 0. In this
case, the f.o.c. yield the following expressions for education and fertility
n∗ (τ q + τ e e∗ ) = γ,
(6)
(e∗ )1−θ
1−θ ∗
qwa
τq
e +
= e.
θ
(1 − q)wm βθ
τ
(7)
The first equation is standard and gives the quantity-quality trade-off faced by parents, and
implies that any shock that reduces optimal education investments, e∗ , results in an increase in
fertility and viceversa. The second equation implicitly defines the optimal education investment
as a function of the two wages and population proportions. This equation captures, as in Galor
and Weil (2000), the fact that education investment in children depends on the way it impacts
the expected wage of the offspring. The main difference with existing work is that investments
in education will depend on the relative returns in the two sectors.
Before we fully solve the model, it is interesting to do some comparative statics with respect
to q and wages. From the two equations above it is straight-forward to show that ∂e∗ /∂q < 0
and ∂n∗ /∂q > 0 , implying that a higher agricultural employment share reduces education and
increases fertility. The intuition for this effect is simply that since education has no value in
the agricultural sector, a higher probability that one’s children work in agriculture reduces the
expected marginal gain of educating an offspring and hence will reduce parents’ incentive to
invest in their education. An increase in the relative wage in agriculture, i.e. a higher value of
the ratio wa /wm , would have the same effect as an increase in agricultural employment.
The full solution to the model requires solving for wages and employment. Since young
adults chose their sector of employment in such a way as to equalize their income, labour
market equilibrium is given by the expression wa = wm h(e), and yields the equilibrium values
of wages and employment. We are interested in the impact of an increase in the price of the
agricultural good, and in the appendix we show that a higher value of p increases the wage rate
in agriculture, leading to a flow of labour into that sector, so that agricultural employment is
q=
ap2
,
ap2 + h(e)
where a ≡ AT /B. A higher price of agricultural goods and a lower level of education increase
employment in agriculture. Note also that if districts differ in the quantity or productivity of
12
their land, they will also differ in their share of employment in agriculture, with a higher A
and/or T (i.e. a higher a) resulting in a higher q.
From equation (7) note that the only magnitude that matters for education decisions is the
ratio of the expected wage in the two sectors, which we denote ω. It is possible to show that in
equilibrium
ω≡
qwa
= ap2 .
(1 − q)wm
The expected relative wage in agriculture is hence increasing in the price of agricultural goods
p.
Suppose the economy faces a price of agricultural goods p and that the resulting fertility
and education decisions are given by n and e. Consider now the introduction of a permanent
tariff on agricultural products at time T that increases the price of agricultural goods to p¿
p. Differentiating the two equilibrium equations, it is straight forward to show that the higher
price will result in an education investment e lower than e and a fertility rate n higher than
n. the former is the result of the decrease in the relative return to education, while the usual
quality-quantity trade-off implies that as parents spend less time in children’s education, they
have more of them. Note also that
d2 e
<0
dadp
and
d2 n
> 0,
dadp
that is, the reduction in education and the increase in fertility are stronger the greater agricultural productivity is. Since a higher a implies that a greater share of population is employed
in agriculture before the price shock, districts which had a high initial employment share in
agriculture will be those experiencing the sharpest changes in our two variables of interest.
3.3
Late education
The results just described concern the comparison between individuals facing an agricultural
price p throughout their lifetime and those facing a price p in all periods. However, for the
generation born at T − 1, parental education investment is given by e , yet their incomes are
determined by the new (higher) price of agricultural goods p.
We thus need to consider the choices of young adults born at time T −1, who face a change in
wages during their life time. These. individuals received from their parents a level of education
e . At the start of period T this cohort makes their occupational choice (between agriculture
and manufacturing) and the tariff is introduced after this choice is made. individuals who are in
the manufacturing sector may revise their education decisions in the light of new wages. rather
13
than having a constant wage throughout their lifetime, they face a wage wm during period T
and are aware that the tariff will lead to an increase in wages next period to wm . This increase
in wages implies that for young adults the cost of education is unchanged but the return has
risen. In the appendix we show that whenever
τe
w
<
0.5b
w
young adults will spend a fraction of time x = 0.5/τ e in education. That is, when the increase
in wages due to the tariff is sufficiently high, the cohort of individuals who are young adults
when the tariff is adopted will experience an increase in the future wage rate and hence invest in
education beyond what their parents gave them. As a result the tariff can lead to a temporary
increase in education, with the human capital of this group of individuals being h(et+2 ) =
βeθt+2 (1 + bxt ) and thus higher than that of both previous and latter generations.
The model hence implies that an increase in the tariff on agricultural goods that raises the
agricultural wage has the following effects:
• parents reduce the educational investment per child, the effect being stronger the larger is
the share of the population employed in agriculture;
• fertility increases, the effect being stronger the larger is the share of the population employed in agriculture;
• in the transition period, young adults invest in education themselves, implying that for
some generations educational attainment may increase.
4
Econometric specification
Inspired by the model above, our empirical specification consists of the following two equations:
Fit = α0 + α1 Si ∗ Mt + ηi + δ1i t + δ2i t2 + it ,
(8)
Eit = β0 + β1 Si ∗ Mt + µi + δ3i t + δ4i t2 + υit ,
(9)
where Fit and Eit are respectively fertility and education in department i at time t, Mt is a
dummy for whether the Méline tariff is in operation at time t and Si is the local share of
employment in cereal production in the year in which the tariff is introduced. This variable
hence acts as a proxy for the capacity for cereal production, and hence the larger Si is, the
14
stronger the effect of the tariff will be. The coefficients δ1i to δ4i capture the impact of districtspecific time trends affecting fertility and education, while ηi and µi are district fixed-effects.
Unified growth theory predicts a trade-off between fertility and education so that the coefficients
α1 and β1 are of opposite sign. The model above implies that the tariff acts a negative shock to
the returns to education, leading to higher fertility and lower education, so that we expect α1 ¿0
and β1 ¡0.
The time structure of the impact of a policy is crucial, as discussed by Wolfers (2006). Although the effect of the tariff on prices is immediate, fertility and education are likely to respond
with a lag because wages may adjust slowly and bearing children and educating them take time,
but also because both variables are affected by social norms resulting from past behaviour that
may slowdown the reaction to policy. We will thus consider two further specifications for each
of our dependent variables. For fertility, the first one takes the form
Fit = α0 + α1 Si ∗ Mt + α2 Si ∗ Mt ∗ Expt + ηi + δ1 t + δ2 t2 + it ,
(10)
where Expt denotes the number of years of exposure to the policy, and we expect the coefficient
α2 to be positive, indicating that households take time to adjust their fertility to the policy. An
alternative specification, based on Wolfers’ analysis of divorce laws, allows for a different impact
of the tariff in different years, that is,
Fit = α0 +
X
αk Si ∗ Mt + ηi + δ1 t + δ2 t2 + it .
(11)
k≥1
This specification allows for greater flexibility when estimating the impact of the policy. It allows,
for example, for the possibility that there is little impact immediately after the introduction of
the tariff while fertility norms adapt to the new regime.
Similarly, we consider two specifications for education which take the form
Eit = β0 + β1 Si ∗ Mt + β2 Si ∗ Mt ∗ Expt + µi + δ3 t + δ4 t2 + υit ,
X
Eit = β0 +
βk Si ∗ Mt + µi + δ3 t + δ4 t2 + υit .
(12)
(13)
k≥1
Recall that the model predicts that for individuals who were mature at the time of the shock, the
tariff increases their educational attainment. We will hence estimate these equations for different
age groups allowing β1 and βk to differ depending on whether we consider the enrolment rates
of children or of young adults.
15
5
The data
Although France has relatively good historical data, the difficulty lies in the unit of observation
that we are interested in: the district or département, which we will term ’department’ through
the paper. These were the regional administrative units at the time, and are still the main
administrative units in France with most of them covering the same areas and having the same
names as in the late 19th century, although the number has slightly increased.
We use three sources to compile our data on education and fertility. The first is the Annuaire
Statistique de la France, from which we have regional data on live births, total population, and
the number of students enrolled in primary education. To create measures of fertility, enrollment
and attendance, we use the census or Recensement Général, which is available for the years 1872,
1876, 1881, 1886, 1891, 1896, 1901, 1906, and 1911, and provides data on various groups of
population by age and gender. Lastly we use data concerning army conscripts. At the time, all
20-year-old men had to report for military service, and a number of individual characteristics,
including educational attainment, were recorded.9 The source are the Compte rendu sur le
recrutement de l’armée, where data on those reporting for military service was collected in a
consistent way at the department level from 1850 up to 1912.
Birth rates by department are defined as the number of live births per 1,000 inhabitants, while
the fertility rate is computed as the ratio of live births to the number of women aged between 15
and 49 in 1,000s. Demographers have raised concerns about a number of observations given in
the census as in certain years the various measures available are not consistent with each other.
Corrections of these data have been proposed to take into account this concern and we use those
to calculate the fertility rate, as proposed by Van de Walle (1974) and Bonneuil (1997).
Our main measure of educational attainment comes from the army data. We focus on the
proportion of potential recruits that can at least read and write. An alternative would have
been to consider the share of those with a primary education degree, but these were still a very
small proportion of the population even in our last sample year (less than 5% of 20-year-old
men have such a degree in 1912). These data have advantages and disadvantages over more
standard measures. Its main advantage is that because education is measured at age 20, it
captures completed education10 and deals with the possibility, not uncommon in that period,
of an individual receiving basic education well into his teens. We use the data for the years
9
These data have been previously used, amongst others, by Banerjee, Duflo, Postel-Vinay, and Watts (2010)
who look at the effect of income shocks on health.
10
Some individuals may study after the age of 20 years is they obtain a higher education degree. However, the
fact that we focus on lower levels of educational attainment makes this possibility irrelevant for our purposes.
16
1872 to 1912, which corresponds to individuals born between 1852 and 1892. The last cohort
that we consider was hence born on the year in which the tariff was introduced, and could have
received primary education over the period 1898 and 1905 (i.e. aged 6 to 13) or latter. That is,
our sample consists of individuals presumably unaffected by the tariff, say those who were 14
or older (i.e. born in 1878 or earlier), and individuals for whom the tariff is likely to have had
an impact in their level of education, those under 13 years of age (i.e. born between 1879 and
1892).
These data have two major drawbacks. The first is that women are not in the sample, and
it is conceivable that the tariff had different effects across the genders. For example, if the tariff
made agriculture a more desirable occupation and if this was largely a male-dominate activity,
girls’ education could have been affected less than boys’. Alternatively, if the tariff had a positive
impact on fertility, this may have kept more girls at home to help with household chores and
caring for younger siblings. A second concern is that the army educational data does not cover
the entire period over which we would expect the tariff to have an impact. For these reasons we
also consider enrolment rates in primary education. Data are available for the overall number
of students enrolled in primary education and for those aged 6 to 13, the difference between the
two being presumably older students.11 The data are available separately for all students, boys
and girls, so we compute both overall and gender-specific enrolment rates. To obtain enrolment
rates for those in the relevant age group we use the population aged 6 to 13, which is available on
census years (1881, 1886, 1891, 1896, 1901, 1906), hence the last observation includes individuals
born in 1900, i.e. 8 years after the tariff was introduced. As discussed above the population
data by age group is not always reliable, and in a number of cases the enrolment rate we obtain
is well over 100%. Since no correction is available for this age group, we simply remove from our
sample the observations that are 101% or higher. As an alternative measure we also compute
enrolment of those aged between 6 and 13 years as a share of the total department’s population.
The number of students enrolled in primary education outside the standard age group (6
to 13 years) can be substantial, amounting to between 35 and 45% of those enrolled in some
departments. We therefore construct two additional measures of enrolment: the first is the
overall number of students enrolled the total population, the second is those enrolled who are
outside the 6-13 age group over the total population.
We start our sample in 1872 and if possible we compile data up to 1913, yielding a 42-year
period with half of the observations pre-dating the Méline tariff and half of them occurring
11
See Grew and Harrigan (1991) for an introduction to the data and Luc (1985) for a discussion on the method
used by the French education ministry to survey the enrolled.
17
after the policy was in place. We exclude from our sample Alsace and parts of Lorraine due
to the annexation by Prussia during the period, as well as Corsica for which there is no data
on agricultural employment, thus reducing our sample to 85 districts. Four observations are
missing for Meurthe et Moselle between 1872 and 1875, as the district was a merge of the two
remaining parts of former districts 54 and 57 that were no longer part of France following the
1870 war. Our sample hence contains at most 3566 observations, all of which are available for
birth rates. For completed education, we have observations on only 40 years, yielding a total
of 3400 observations12 , while for fertility and enrolment rates the quinquennial availability of
censuses reduces our sample to around 500 observations.
Our policy variable is the interaction between a dummy for the Méline tariff and a measure
of the importance of cereal production in the department’s economy in 1892. Data on the share
of employment in cereal production are not available, hence we use as a proxy the product of
the share of agricultural employment in total employment in 1892 and the share of the value
of cereal production in total agricultural production in 1892, i.e. the last year before the tariff
could have an impact. The data concerning these two variables comes from Van de Walle (1974)
and Bonneuil (1997).
The dummy variable ‘Méline’ takes the value 0 up to 1892 and the value 1 from 1893 onwards,
1893 being the first year in which we could observe a change in fertility or education. As discussed
the time structure of the effect of the policy is of crucial importance, as this variable can have
different effects depending on how long the policy has been in operation. We will thus use the
variable ‘exposure’ to measure the number of years that the policy has been in place, and will
also allow for differential impacts every three or five years.
Table 1 presents some descriptive statistics. Meline is the interaction term between the share
of employment in cereal production and the dummy taking a value of one from 1893 onwards
and zero for earlier years. As we can see in the table, cereal production was an important
activity in France. Its share of employment averages 14.6%, and varied between 26% and 0.07%,
with Lot, Tarn et Garonne and Dordogne being the districts with the highest shares and Seine
that with the lowest. Note, however, that not all districts with a low employment share in
cereals were rich, urban regions. The third lowest share is that of Boûches du Rhone, at 3.5%, a
relatively poor region but whose climate is not suitable for cereal production. Figure 3 represents
the spatial distribution of the share of employment in cereal production. Both birth rates and
12
The two missing years are 1913, for which we have no data, and 1878, which we have removed from the
sample because of inconsistencies for various departments in the data for that year and previous and subsequent
observations.
18
Figure 3: Employment share in local production across French districts in 1892
fertility rates are high although declining throughout the period, when the average in the sample
being 92 children per thousand women. The literacy of men measured at age 20 is high, with
an average of 91%, but its variance is large too with some districts exhibiting rates of 56% and
others having a fully literate population.
6
6.1
Empirical results
Fertility
Table 2 reports the results for birth rates. The first column simply includes a 0-1 dummy
starting in 1893 which is interacted with the share of employment in cereal production, as well
as a department-specific linear time trend. The effect of the dummy is positive and highly
significant, indicating that protectionism increased birth rates in those departments with a
higher share of cereal employment. The second column considers the impact of the time during
which the policy has been in place and finds that the effect grows over time. Column 3 presents
the most flexible specification, based on equation (11), which allows for differential effects every
three years, and indicates that the effect on birth rates increases over time, rapidly in the first
decade and more slowly afterwards. This seems to imply that households adapted their fertility
19
gradually in response to the change in the relative return to education. The next three columns
estimate those specifications including both a linear and a quadratic district-specific time trend.
coefficients have the same sign and significant, and are somewhat larger.
Table 3 reports the same specifications using as the dependent variable fertility rates, where,
because census data reporting the number of females of child-bearing age is only available every
five years, we have interpolated the female population to obtain annual fertility rates. Results
using only census years are reported the appendix. The results are consistent with those obtained
with birth rates: the interaction between the tariff and cereal production has a positive and
significant coefficient and so does the term in which we take into account the number of years of
exposure to the policy. Columns 3 and 6 report regressions based on equation (11). Interestingly,
we find that the effect increases dramatically after the first three years, as well as between year 6
and year 9, to then stabilize (even decline in the case of a linear trend). The magnitude of these
effects is large. In a department with 25% of the population employed in cereal production, i.e.
the highest shares that we observe, the introduction of the tariff increased the fertility rate by
8.4 children per 1000 women, and evaluated at the sample mean of cereal employment (15 %),
by 5.1 children per 1000 women. These figures are equivalent to 50% and 29% of the standard
deviation of fertility. As we have discussed, this was a period of declining fertility rates and it
is interesting to compare the impact of the policy with that of the time trend, since the former
offset the decline in fertility that had been taking place since the late 18th century. Using the
formulation in column 3 (since the linear trend is easier to interpret), the estimated coefficients
imply that, for a district with average cereal employment, fertility rates increased following the
introduction of the tariff and reached their 1892 rate only 14 years after its introduction. In
other words, the tariff implied a 14-year delay in the reduction of fertility.
6.2
Education
Consider now the effect on education. Table 4 presents the results for enrolment rates, defined
as the number of students registered in primary education over the relevant age group (6 to 13
year-olds). We report results for all children, for boys only and for girls only since, as we have
argued above, the effect could be different across the sexes.
For each dependent variable we use two specifications, one simply including the Méline tariff
interacted with cereal employment, and another that allows for a different effect in years four,
nine, and 14 after that tariff’s introduction (these being the census years for which we have data
on population by age). The last observation is hence for 1906 and includes individuals born
between 1893 and 1900, i.e. up to 8 years after the tariff was introduced. The coefficient on our
20
variable of interest is negative and significant in all specifications, and we find no statistically
significant difference between boys and girls. The magnitude of the effect is substantial : for a
15% employment rate in cereal production, the tariff reduces enrolment rates by 3.4 percentage
points, which amounts to almost 60 percent of the standard deviation. These effects are very
large when we compare them to the evolution of enrolment rates over time: over the decade
prior to the introduction of the tariff, the enrolment rate increase by only 1.4 percentage points
in France. When we allow for different effects across time, we find that the strongest impact
occurs nine years after the introduction of the tariff, while the effect is insignificant for year 14.
One problem with the data on enrolment rates is that primary education registries indicate
that a substantial fraction of students are older than 13. Our model indicates that the impact of
the tariff on education may vary depending on the individual’s age. We hence consider separately
the number of pupils aged 6 to 13 and those aged over 13. Since we do not have the age range
for the latter group, we construct enrolment rates defined as the number of pupils over the total
population. The results are reported in table 5. The first two columns consider pupils of all ages
and find a negative effect except for the last census year (1911). The 1911 census does not report
population by detailed age groups, column three runs the same regression dropping that sample
year. Columns 3 and 4 report results for the 6 to 13 age group. The tariff reduces enrolment
of this age group in high cereal-producing districts, with the effect increasing over time and
being significant in all periods. The last two columns of the table examine the enrolment rate of
those older than 13, and find an increase in educational attainment in the first decade after the
introduction of the tariff. These results support the idea that income effects led young adults
to increase their human capital after the policy was introduced.
To further explore differences in educational attainment across age groups, we consider the
data concerning military recruits, which has the advantage of giving us information on completed
education. The disadvantage of these data is that it consists mainly of individuals before the
introduction of the tariff. The first specification allows for a different effect depending on the time
over which the policy has been in operation. The years immediately following the introduction
of the tariff (years 1 to 3) exhibit a positive coefficient, indicating that for those young men
turning 20 between 1893 and 1895 educational attainment increased. For the next two groups
(men turning 20 between 1896 and 1901) we find a negative but insignificant coefficient, probably
the result of offsetting positive and negative impacts. For all younger generations, the coefficient
on our policy variable is negative and becomes larger over time. For those born in 1893-94, the
impact of the policy amounts to a reduction of the fraction of young men who can at least
21
read and write of 5 percentage points in a district with the average employment share in cereal
production. In contrast, the positive effect for older generations is weak, increasing educational
attainment by the 0.6 percentage points for a district with average employment.
Column 2 examines the effect of exposure and finds a negative coefficient, as expected. The
next six columns consider groups of different ages. We start by examining the effect of the tariff
on individuals that were 9 or 10 when it was introduced, and then consider those that were
11 or 12 and so on up to 20. For the first two groups we find a negative effect of the tariff,
the coefficient is insignificant for those younger than 14 and 16, and becomes positively and
significant for those that were older in 1892.
7
Conclusions
TO BE WRITTEN
8
Appendix 1
This appendix derives some of the results reported in section 3.
TO BE WRITTEN
9
Appendix 2
This appendix gives further details on the data.
TO BE WRITTEN
22
References
Accominotti, O. and M. Flandreau (2008). Bilateral treaties and the Most-Favored-Nation
clause. the myth of trade liberalization in the 19th century. World Politics 60, 147–188.
Augé-Laribé, M. (1950). La politique agricole de la France de 1880 a 1940. Paris: Presses
universitaires de France.
Bairoch, P. (1972). Free trade and european economic development in the 19th century. European Economic Review 3, 211–245.
Banerjee, A., E. Duflo, G. Postel-Vinay, and T. Watts (2010). Long-run health impacts of
income shocks: Wine and phylloxera in nineteenth-century France. Review of Economics
and Statistics 92 (4), 714–728.
Bardet, J.-P. and H. Le Bras (1988). Histoire de la population franaise, Volume 3, Chapter
La chute de la fcondit, pp. 351–402. Paris: Presses Universitaires de France.
Becker, G. S. (1960). Demographic and Economic Change in Developed Countries, Chapter
An Economic Analysis of Fertility, pp. 209–249. Princeton, N.J.: Princeton University
Press.
Becker, G. S. and N. Tomes (1976). Child endowments and the quantity and quality of children. Journal of Political Economy 84 (4), 143–162.
Becker, S. O., F. Cinnirella, and L. Woessmann (2010). The trade-off between fertility and education: evidence from before the demographic transition. Journal of Economic Growth 15,
177–204.
Black, S. E., P. J. Devereux, and K. G. Salvanes (2005). The more the merrier? the effect of
family size and birth order on childrens education. Quarterly Journal of Economics 120 (2),
669–700.
Blayo, Y. and L. Henry (1975). La population de la France de 1840 á 1860. Population 30 (H–
S), 71–122.
Bleakley, H. and F. Lange (2009). Chronic disease burden and the interaction of education,
fertility, and growth. Review of Economics and Statistics 91 (1), 52–65.
Bonneuil, N. (1997). Transformation of the French Demographic Landscape, 1806–1906. Oxford: Clarendon Press.
Chaunu, P. (1972). Malthusianisme dmographique et malthusianisme conomiques. Annales:
Economies, societes, Civilisations 26, 1–19.
23
Cummins, N. (2013). Marital fertility and wealth during the fertility transition : Rural France,
1750–1850. economic history review 66 (2), 449–476.
Dormois, J.-P. (2009). La défense du travail national: L’incidence du protectionnisme sur
l’industrie en Europe (1870–1914). Paris: Presse Universitaire Paris Sorbonne.
Dupaquier, J. (1988). Histoire de la population franaise. Paris: Presses universitaires de
France.
Easterlin, R. A. (1976). The conflict between aspirations and resources. Population and Development Review 2 (3), 417–425.
Federico, G. and K. Persson (2007). The New Comparative Economic History: Essays in
Honor of Jeffrey G. Williamson, Chapter Market integration and convergence in the world
wheat market, 1800-2000, pp. 87114. Cambridge, MA: MIT Press.
Galor, O. and O. Moav (2002). Natural selection and the origin of economic growth. Quarterly
Journal of Economics 117 (4), 1133–1191.
Galor, O. and D. N. Weil (1999). From malthusian stagnation to modern growth. American
Economic Review 89 (2), 150–154.
Galor, O. and D. N. Weil (2000). Population, technology, and growth: From malthusian
stagnation to the demographic transition and beyond. American Economic Review 90 (4),
806–828.
Golob, E. O. (1944). The Méline tariff: French agriculture and nationalist economic policy.
New York: Studies in History, economics and public law–AMS press.
Grew, R. and P. J. Harrigan (1991). School, State and Society, The Growth of Elementary
Schooling Nineteenth-Century France–A Quantitative Analysis. Ann Arbor: Clarendon
Press.
Guinnane, T. W. (2011). The historical fertility transition: A guide for economists. Journal
of Economic Literature 49 (3), 589–614.
Harley, K. (1988). Ocean freight rates and productivity, 17401913: the primacy of mechanical
invention reaffirmed. Journal of Economic History 48 (4), 851–876.
Jacks, D. P. (2006). New results on the tariff-growth paradox. European Review of Economic
History 10, 205–230.
Kindleberger, C. P. (1950). Group behavior and international trade. Journal of Political Economy 59 (1), 30–46.
24
Levasseur, E. (1911). Histoire du commerce de la France. Paris: Librairie nouvelle de droit et
de jurisprudence.
Luc, J.-N. (1985). La Statistique de l’Enseignement Primaire 19e-20e Si’ecle, Politique et
Mode d’Emploi. Paris: INRP Economica.
North, D. C. (1958). Ocean freight rates and economic development 17501913. Journal of
Economic History 18 (4), 537–555.
Nye, J. V. (2007). War, wine, and taxes: the political economy of Anglo–French trade; 1689–
1900. Princeton: Princeton University Press.
O’Rourke, K. H. (1997). The european grain invasion, 1870–1913. Journal of Economic History 57 (4), 775–801.
O’Rourke, K. H. (2000). Tariffs and growth in the late 19th century. Economic Journal 110,
456–483.
Rosenzweig, M. R. and K. I. Wolpin (1980). Testing the quantity-quality fertility model: The
use of twins as a natural experiment. Econometrica 48 (1), 227–240.
Rosenzweig, M. R. and J. Zhang (2009). Do population control policies induce more human
capital investment? twins, birthweight, and chinas one child policy. Review of Economic
Studies 76 (3), 1149–1174.
Schultz, P. T. (1985). Changing world prices, womens wages, and the fertility transition:
Sweden, 1860–1910. Journal of Political Economy 93 (6), 1126–1154.
Tena-Junguito, A., M. Lampe, and F. Tâmega Fernandes (2012). How much trade liberalization was there in the world before and after cobden-chevalier? Journal of Economic
History 72 (3), 708–740.
Uebele, M. (2011). National and international market integration in the 19th century: Evidence from comovement. Explorations in economic history 48, 226–242.
Van de Walle, E. (1974). The female population of France in the nineteenth century: A
reconstruction of 82 departements. Princeton: Princeton University Press.
Van de Walle, E. (1980). Family and Sexuality in French History, Chapter Motivations and
Technology in the Decline of French Fertility. Philadelphia: University of Pennsylvania
Press.
Weir, D. R. (1984). Life under pressure: France and England, 1670–1870. Journal of Economic
History 44 (1), 27–47.
25
Wolfers, J. (2006). Did unilateral divorce laws raise divorce rates? a reconciliation and new
results. American Economic Review 96 (5), 1802–1820.
Wrigley, A. E. (1985a). The fall of marital fertility in nineteenth century France: Exemplar
or exception? (part I). European Journal of Population 1 (1), 31–60.
Wrigley, A. E. (1985b). The fall of marital fertility in nineteenth century France: Exemplar
or exception? (part II). European Journal of Population 1 (2–3), 141–177.
26
10
Tables
Table 1: Descriptive statistics
Employment share in cereal
Meline
Birth rate
Fertility rate
Literacy rate
Enrolment rate, all
Enrolment rate, 6–13
Enrolment rate, not 6–13
N
85
85
3566
3566
3566
680
595
595
Mean
.1485415
.0637092
22.46258
93.97346
.8991501
14.31238
11.34968
2.962695
27
Standard Deviation
.0611807
.0837492
3.940087
17.19076
.0850998
2.250529
1.515979
1.073554
Min
.0006763
0
13.80399
54.88631
.4431624
6.064898
5.171001
-1.379808
Max
.2614087
.2614087
35.4283
175.0147
.9973357
23.21676
17.05617
7.743848
Table 2: Birth rate
(1)
(2)
(3)
Meline Exposure Dynamic
4.395∗∗∗
(0.617)
Meline
(4)
Meline
4.404∗∗∗
(0.625)
0.534∗∗∗
(0.0670)
Exposure
(5)
Exposure
(6)
Dynamic
2.041∗∗∗
(0.185)
Years 1-3
2.646∗∗∗
(0.547)
3.669∗∗∗
(0.399)
Years 4-6
7.449∗∗∗
(0.749)
9.283∗∗∗
(0.689)
Years 7-9
9.264∗∗∗
(0.924)
12.09∗∗∗
(1.029)
Years 10-12
11.84∗∗∗
(1.070)
15.84∗∗∗
(1.404)
Years 13-15
11.52∗∗∗
(1.223)
16.89∗∗∗
(1.887)
Years 16-18
13.61∗∗∗
(1.351)
20.53∗∗∗
(2.467)
Years 19-21
12.58∗∗∗
(1.450)
21.24∗∗∗
(2.932)
Linear trend *departement
Quadratic trend *departement
Adjusted R2
Observations
∗
p < 0.05,
∗∗
p < 0.01,
∗∗∗
Yes
Yes
Yes
Yes
Yes
Yes
No
0.880
3566
No
0.886
3566
No
0.895
3566
Yes
0.900
3566
Yes
0.905
3566
Yes
0.909
3566
p < 0.001; Standard errors in parentheses
Notes: (1) The period of estimation is 1872-1913; (2) Residuals are clustered at the departement level.
28
Table 3: Fertility rate
(1)
(2)
(3)
Meline Exposure Dynamic
33.71∗∗∗
(4.429)
Meline
(4)
Meline
33.73∗∗∗
(4.484)
1.812∗∗∗
(0.335)
Exposure
(5)
Exposure
(6)
Dynamic
23.46∗∗∗
(1.832)
Years 1-3
3.817
(3.348)
14.42∗∗∗
(2.122)
Years 4-6
30.65∗∗∗
(4.446)
49.67∗∗∗
(3.558)
Years 7-9
69.81∗∗∗
(6.952)
99.19∗∗∗
(7.641)
Years 10-12
88.18∗∗∗
(7.098)
129.8∗∗∗
(9.582)
Years 13-15
80.17∗∗∗
(8.308)
136.0∗∗∗
(12.71)
Years 16-18
60.95∗∗∗
(6.781)
132.9∗∗∗
(13.57)
Years 19-21
29.47∗∗∗
(7.595)
119.5∗∗∗
(15.39)
Linear trend *departement
Quadratic trend *departement
Observations
Adjusted R2
∗
p < 0.05,
∗∗
p < 0.01,
∗∗∗
Yes
Yes
Yes
Yes
Yes
Yes
No
3566
0.603
No
3566
0.592
No
3566
0.700
Yes
3566
0.619
Yes
3566
0.684
Yes
3566
0.722
p < 0.001. Standard errors in parentheses
Notes: (1) The period of estimation is 1872-1913; (2) Residuals are clustered at the departement level.
29
Table 4: Enrolment rate of pupils aged 6-13 over number of cohorts 6-13
(1)
(2)
(3)
(4)
(5)
(6)
All children All children
Boys
Boys
Girls
Girls
Meline
-53.60∗∗∗
-50.21∗∗∗
-56.73∗∗∗
(4.977)
(5.610)
(4.868)
Year 4
-45.95∗∗∗
(5.240)
-42.06∗∗∗
(5.816)
-49.60∗∗∗
(5.171)
Year 9
-74.19∗∗∗
(5.961)
-70.71∗∗∗
(6.717)
-77.16∗∗∗
(5.847)
Year 14
-56.53∗∗∗
(7.971)
-50.47∗∗∗
(8.604)
-61.92∗∗∗
(8.299)
Linear trend
Adjusted R2
Observations
∗
p < 0.05,
∗∗
Yes
0.301
595
p < 0.01,
∗∗∗
Yes
0.348
595
Yes
0.245
595
Yes
0.298
595
Yes
0.316
595
Yes
0.350
595
p < 0.001. Standard errors in parentheses
(1) Enrolment and schooling population are available for years 1876, 1881, 1886, 1891, 1896, 1901, 1906;
(2) The number of pupils aged 6-13 is not available in 1911;
(3) Residuals are clustered at the departement level.
Table 5: Enrolment rate of pupils over district population
(1)
(2)
(3)
(4)
(5)
All ages All ages All ages w/o 1911
6–13
6–13
-5.778∗∗∗
-7.221∗∗∗
(0.521)
(0.635)
Meline
(6)
Not 6–13
0.204
(0.419)
(7)
Not 6–13
Year 4
-6.940∗∗∗
(0.638)
-6.940∗∗∗
(0.644)
-7.042∗∗∗
(0.662)
0.102
(0.470)
Year 9
-10.75∗∗∗
(1.024)
-10.75∗∗∗
(1.034)
-9.970∗∗∗
(0.898)
-0.777
(0.479)
Year 14
-14.09∗∗∗
(1.442)
-14.09∗∗∗
(1.456)
-11.82∗∗∗
(1.172)
-2.271∗∗∗
(0.611)
Year 19
-16.18∗∗∗
(1.759)
Linear trend *dpt
Adjusted R2
Observations
∗
p < 0.05,
∗∗
p < 0.01,
Yes
0.365
680
∗∗∗
Yes
0.464
680
Yes
0.440
595
Yes
0.342
595
Yes
0.396
595
Yes
0.318
595
p < 0.001. Standard errors in parentheses
(1) Enrolment and schooling population are available for years 1876, 1881, 1886, 1891, 1896, 1901, 1906, 1911;
(2) The number of pupils aged 6-13 is not available in 1911;
(3) Column 1 & 2 differs from others by the the inclusion of 1911 data;
(4) Residuals are clustered at the departement level.
30
Yes
0.367
595
Years 1-2
Years 3-4
Years 5-6
Years 7-8
Years 9-10
Years 11-12
Years 13-14
Years 15-16
Years 17-18
Years 19-20
Meline
Linear trend*dpt
Quadratic t.*dpt
Observations
Adjusted R2
Table 6: Educational achievement of 20-year old males
(1)
(2)
(3)
(4)
(5)
(6)
Dynamic Exposure
Age l0
Age l2
Age l4
Age l6
0.0263
(0.0161)
-0.00267
(0.0259)
-0.0557
(0.0362)
-0.115∗
(0.0520)
-0.163∗
(0.0620)
-0.209∗∗
(0.0788)
-0.320∗∗
(0.0981)
-0.357∗∗
(0.116)
-0.425∗∗
(0.138)
-0.507∗∗
(0.161)
-0.0225∗∗ -0.0586∗∗ -0.0307∗ -0.0101
0.0278
(0.00662) (0.0178) (0.0129) (0.0150) (0.0183)
Yes
Yes
3400
0.882
Yes
Yes
3400
0.878
Yes
Yes
3400
0.877
Yes
Yes
3400
0.877
Yes
Yes
3400
0.877
Yes
Yes
3400
0.877
(7)
Age l8
0.0704∗∗∗
(0.0205)
Yes
Yes
3400
0.878
Notes:
(1) Standard errors in parentheses;
∗
p < 0.05,
∗∗
p < 0.01,
∗∗∗
p < 0.001
(2) Information on educational achievement is available yearly between 1872 and 1912;
(3) Headings of columns (3)–(8) mention the maximum age of the boys when Meline was introduced.
For example Age 18 means boys who were aged 18 or less when Meline was adopted, i.e. born after 1874;
(4) dpt stands for departement;
(5) Residuals are clustered at the departement level.
31
Table 7: Fertility rate - census years only
(1)
(2)
(3)
Meline Exposure
Dynamic
54.91∗∗∗
(5.541)
Meline
1.548∗∗∗
(0.418)
Exposure
Year 4
18.77∗∗∗
(4.600)
Year 9
94.57∗∗∗
(9.012)
Year 14
91.65∗∗∗
(10.23)
Year 19
25.49∗∗
(8.547)
Linear trend *departement
Adjusted R2
Observations
∗
p < 0.05,
∗∗
p < 0.01,
∗∗∗
Yes
0.557
763
Yes
0.520
763
Yes
0.685
763
p < 0.001. Standard errors in parentheses
Data: Every 5 years between 1872-1911 except for 1872 and 1876 for district 54
32